Abstract

   The extracellular matrix (ECM) undergoes substantial changes during
   prostate cancer (PCa) progression, thereby regulating PCa growth and
   invasion. Herein, a meta‐analysis of multiple PCa cohorts is performed
   which revealed that downregulation or genomic loss of ITGA1 and ITGA2
   integrin genes is associated with tumor progression and worse
   prognosis. Genomic deletion of both ITGA1 and ITGA2 activated
   epithelial‐to‐mesenchymal transition (EMT) in benign prostate
   epithelial cells, thereby enhancing their invasive potential in vitro
   and converting them into tumorigenic cells in vivo. Mechanistically,
   EMT is induced by enhanced secretion and autocrine activation of TGFβ1
   and nuclear targeting of YAP1. An unbiased genome‐wide co‐expression
   analysis of large PCa cohort datasets identified the transcription
   factor TEAD1 as a key regulator of ITGA1 and ITGA2 expression in PCa
   cells while TEAD1 loss phenocopied the dual loss of α1‐ and
   α2‐integrins in vitro and in vivo. Remarkably, clinical data analysis
   revealed that TEAD1 downregulation or genomic loss is associated with
   aggressive PCa and together with low ITGA1 and ITGA2 expression
   synergistically impacted PCa prognosis and progression. This study thus
   demonstrated that loss of α1‐ and α2‐integrins, either via
   deletion/inactivation of the ITGA1/ITGA2 locus or via loss of TEAD1,
   contributes to PCa progression by inducing TGFβ1‐driven EMT.

   Keywords: ECM, EMT, epithelium, integrin, prostate
     __________________________________________________________________

   Cell‐extracellular matrix interactions are critical regulators of
   cancer cell growth and invasion. This study revealed depletion of two
   collagen‐binding receptors, α1‐ and α2‐integrins, in aggressive forms
   of prostate cancer. Loss of α1‐ and α2‐integrins triggered an autocrine
   secretion of TGFβ leading to EMT and elevated metastatic capacity.
   TEAD1 is identified as a key transcription factor regulating ITGA1 and
   ITGA2.

   graphic file with name ADVS-11-2305547-g005.jpg

1. Introduction

   Prostate cancer (PCa) is the second most commonly diagnosed cancer and
   the fifth leading cause of cancer deaths in men in the Western world.^[
   [50]^1 ^] PCa progression is accompanied by gradual loss of epithelial
   organization and reorganization of the associated extracellular matrix
   (ECM) in the affected prostate glands. Heterodimeric integrins are the
   major family of ECM receptors. Integrin ligation to the ECM not only
   mediates adhesion but also leads to assembly of large signaling
   platforms that convey the microenvironmental cues to regulate cell
   growth, survival and migration.^[ [51]^2 ^] Several studies have
   implicated altered integrin functions in PCa pathogenesis.^[ [52]^3 ,
   [53]^4 ^]

   Loss of PCa cell polarity coincides with reduced expression of laminin,
   a central basement membrane component, and disassembly of
   α6β4‐integrin‐mediated hemidesmosomal laminin adhesions.^[ [54]^5 ,
   [55]^6 , [56]^7 ^] PCa progression is associated with increased
   expression of collagen that is the major constituent of stromal ECM.
   α1β1‐ and α2β1‐integrins are collagen‐binding integrins that may also
   interact with laminins and are expressed in prostate epithelial
   cells.^[ [57]^8 , [58]^9 ^] α2β1‐integrins have been reported to
   promote tumorigenesis, particularly metastatic invasion to the bone.^[
   [59]^4 ^] However, other studies have observed tumor suppressive
   activity and/or loss of α2β1‐integrin expression during PCa
   progression.^[ [60]^8 , [61]^10 , [62]^11 ^] To our knowledge, the
   potential role of α1β1‐integrins or synergistic role of the neighboring
   ITGA1 and ITGA2 genes, both encoding for collagen‐binding integrin
   α‐subunits, have not been addressed in human prostate tumorigenesis and
   PCa progression.

   Here we have performed a meta‐analysis consisting of several large
   cohorts of PCa patients and revealed integrin signaling as the most
   significantly affected pathway associated with PCa progression.
   Remarkably, genomic co‐deletion or downregulation of ITGA1 and ITGA2 in
   PCa tumors was significantly associated with disease progression. We
   hence sought to investigate the potential role of α1β1‐ and
   α2β1‐integrins in promoting tumorigenicity of PCa. Functional studies
   showed that both α1‐ and α2‐integrins independently contribute to
   collagen adhesion and glandular morphogenesis of prostate epithelial
   cells. Simultaneous loss of α1‐ and α2‐integrins induced transformation
   of benign prostate epithelial cells, leading to increased invasive
   properties in vitro and metastatic capacity in vivo. Mechanistically,
   this effect was mediated by induced secretion and activation of
   autocrine TGFβ1 that promoted epithelial‐to‐mesenchymal transition
   (EMT) and nuclear translocation of YAP1. YAP1 transcriptional
   co‐regulator and binding partner of TEAD1 was found to control ITGA1
   and ITGA2 expression. Strikingly, loss of TEAD1 phenocopied α1α2‐dKO
   cells as they were transformed with increased metastatic capacity in
   vitro and in vivo. Therefore, our work has deciphered a novel mechanism
   by which TEAD1 regulates α1‐ and α2‐integrin expression and thereby
   exerts tumor suppressive functions while loss of TEAD1 and/or
   ITGA1/ITGA2 in PCa genomes were found to correlate with disease
   progression and aggressiveness.

2. Results

2.1. A Meta‐Analysis Reveals Prevalent Downregulation of Integrin Signaling
Pathway Genes in PCa Development

   The roles of androgens and AR‐signaling are well‐documented in
   metastatic PCa. However, several other cellular signaling cascades may
   contribute to PCa development and progression. To comprehensively
   dissect major signaling pathways involved in PCa, we performed a
   meta‐analysis by incorporating nine clinical PCa expression profiling
   data. The meta‐analysis identified a total of 1284 downregulated and
   610 upregulated genes over PCa tumor progression (Figure [63] 1a).
   Differentially expressed genes (DEGs) were subsequently subjected to
   PANTHER Pathway analysis for an over‐representation test, which
   remarkably, revealed integrin signaling pathway, consisting of 191
   genes, as the most significant cellular pathway deregulated during PCa
   progression (Figure [64]1b). Notably, the meta‐analysis identified a
   total of 55 DEGs from the integrin signaling pathway and most of them
   were downregulated upon PCa development (Figure [65]S1a, Supporting
   Information).

Figure 1.

   Figure 1
   [66]Open in a new tab

   Integrin signaling pathway and the copy number loss/del of ITGA1 or
   ITGA2 are associated with PCa severity and prognosis. a) Differentially
   expressed genes identified from a meta‐analysis consisting of nine
   independent PCa cohorts. Up‐ and downregulated genes were highlighted
   in orange and blue, respectively. b) PANTHER pathway ontology analysis
   identified the Integrin signaling pathway as the most affected pathway
   in PCa. c) ITGA1 and ITGA2 were determined with the highest frequencies
   of deep copy loss in PCa among other cancer types (ovarian – ov,
   esophageal – esca, bladder – blca, stomach – stca, cervical – cesc,
   head and neck – hnsc, breast – brca, lung – lnsc and luad,
   pheochromocytoma and paraganglioma – pcpg, colorectal ‐coadread,
   uterine corpus endometrial – ucec and brain – gbm). d) ITGA1 or e)
   ITGA2 expression in prostate tumors with diploid or copy number
   loss/del of ITGA1 or ITGA2 (left panel), and Pearson correlation
   between ITGA1 or ITGA2 mRNA expression and copy number changes (right
   panel). P values determined by the Mann‐Whitney U test or from the
   Pearson correlation coefficient. f) The fraction of PCa tumors
   harboring ITGA1 or ITGA2 copy number loss/del was elevated in
   metastasis when compared with primary PCa. P values were examined by
   the Fisher's exact test. g,h) Top gene set depleted in PCa tumors with
   ITGA1 g) or ITGA2 h) copy number loss/del versus diploid in the TCGA
   PCa cohort. NES, normalized enrichment score. i,j) Forest plots
   displaying the meta‐analysis of the hazard ratio estimates of ITGA1 i)
   or ITGA2 j) for the biochemical recurrence in multiple PCa cohorts. The
   horizontal error bars represent the 95% CI with the measure of center
   as HR. The HR and 95% CI were presented in the form of natural
   logarithm (ln). P values were calculated by the two‐way Fixed‐Effects
   Model. k,l) Kaplan Meier plots indicate increased risk for metastasis
   in PCa patients with tumors harboring ITGA1k) or ITGA2 l) copy number
   loss/del. P‐values were assessed by log‐rank tests.

2.2. α1‐ and α2‐Integrins Genomic Loss or Downregulation Correlates with PCa
Progression

   Previous studies have shown that genes contributing to tumorigenesis
   are expected to manifest association between DNA copy number
   alterations and RNA expression levels.^[ [67]^12 ^] We therefore sought
   to identify potential causal genes from the Integrin signaling pathway
   by the joint analysis of gene copy number alterations and expression
   levels. We compared the copy number profiles of differentially
   expressed genes from the Integrin signaling pathway in a pan‐cancer
   analysis from the TCGA (n = 10967), including PCa. The result
   pinpointed ITGA1 and ITGA2 as the two genes with the highest
   proportions of deep loss in PCa compared to other cancer types
   (Figure [68]1c).

   We assessed whether the genomic alterations of ITGA1/ITGA2 display any
   clinical impacts on PCa, and thus investigated potential correlation
   between ITGA1/ITGA2 copy number loss and their expression levels in
   human prostate tumors. The results showed that ITGA1/ITGA2 expression
   levels were dramatically decreased in PCa patient tumors with
   ITGA1/ITGA2 copy number loss/deletion (loss/del) in two independent
   clinical PCa cohorts (Figure [69]1d,e; Figure [70]S1b,c, Supporting
   Information). Next, we examined whether ITGA1/ITGA2 copy loss directly
   correlates with PCa metastasis in two independent cohorts of PCa
   patients with primary and metastatic tumors. This analysis revealed
   that the proportion of ITGA1/ITGA2 copy number loss/del was
   substantially higher in PCa patients with metastasis compared to
   patients with localized tumors (Figure [71]1f; Figure [72]S1d,
   Supporting Information). These findings suggest a potential functional
   relevance of ITGA1/ITGA2 copy loss/del in PCa progression. We hence
   conducted a gene set enrichment analysis (GSEA) using the genome‐wide
   expression profiling data in the TCGA cohort to identify gene sets that
   are enriched in phenotypes of diploid and ITGA1/ITGA2 loss/del,
   respectively. These results demonstrated that “CHANDRAN Metastasis DN”,
   a gene signature representing genes downregulated in metastatic versus
   nonmetastatic prostate carcinoma, was identified as the top enriched
   gene set across all gene sets depleted in PCa tumors with ITGA1/ITGA2
   loss/del compared to diploid tumors (Figure [73]1g,h). Interestingly,
   we also found that several cell cycle‐related pathways including E2F
   targets and G2M checkpoint were highly enriched in PCa tumors with
   ITGA1/ITGA2 loss/del (Figure [74]S1e–h, Supporting Information),
   further implicating the functional role of ITGA1/ITGA2 loss/del in PCa
   tumorigenesis and metastasis.

   We further evaluated the clinical impact of copy number alterations of
   ITGA1/ITGA2 on PCa progression and therefore conducted the Kaplan‐Meier
   survival analysis to examine the association of ITGA1/ITGA2 copy number
   loss/del and patient prognosis in multiple independent PCa cohorts. The
   results displayed that PCa patients with ITGA1/ITGA2 copy number
   loss/del were associated with shorter biochemical recurrence‐free
   survival in three distinct PCa cohorts (Figure [75]S1i–n, Supporting
   Information). To validate the robustness of the association, we
   additionally carried out a meta‐analysis to systematically review,
   interpret and estimate the overall association between ITGA1/ITGA2 copy
   number loss/del and PCa patient survival outcomes. This analysis
   demonstrated that PCa patients with ITGA1 copy number loss/del was
   significantly associated with shorter biochemical recurrence‐free
   survival (P = 3.07e‐05, ln(HR) = 0.74; 95%, ln(CI): 0.39–1.09)
   (Figure [76]1i). Similar association was observed for PCa tumor with
   ITGA2 copy number loss/del (P = 1.26e‐05, ln(HR) = 0.77; 95% ln(CI):
   0.42– 1.11) (Figure [77]1j). Having demonstrated the high proportion of
   ITGA1/ITGA2 copy loss/del in metastatic PCa tumors overrepresenting
   metastasis pathways (Figure [78]1f–h), we proceeded to investigate the
   effect of ITGA1/ITGA2 copy number loss/del on patient metastatic
   progression outcomes. The result strongly indicated that ITGA1/ITGA2
   copy number loss/del in PCa patients was associated with increased
   metastatic risk (Figure [79]1k,l). Taken together, these data
   demonstrated that genomic loss/depletion of ITGA1/ITGA2 is accompanied
   with their decreased expression and is significantly associated with
   metastatic PCa progression, thereby suggesting a potential prognostic
   value for the ITGA1/ITGA2‐status in PCa risk prediction.

   Given that the expression levels of ITGA1/ITGA2 were downregulated in
   PCa tumors with copy number loss/del (Figure [80]1d,e;
   Figure [81]S1b,c, Supporting Information), and that ITGA1/ITGA2 genomic
   alterations in turn demonstrated clinical impacts on PCa development
   and progression (Figure [82]1i–l; Figure [83]S1i–n, Supporting
   Information), we next explored whether lower mRNA levels of ITGA1/ITGA2
   correlated with human PCa progression in the clinical setting and thus
   analyzed multiple independent PCa cohorts. The results indicated that
   ITGA1/ITGA2 were reproducibly found to be downregulated in prostate
   tumor tissues compared to normal prostate glands (Figure [84] 2a–f;
   Figure [85]S2a,b,d,e, Supporting Information). In particular, we
   observed that the mRNA levels of ITGA1/ITGA2 were significantly
   decreased in metastatic PCa samples (Figure [86]2c,f; Figure [87]S2a–f,
   Supporting Information). To further validate this finding, we analyzed
   a tissue microarray consisting of matched normal versus cancer samples
   from 242 local PCa patients (Table [88]S6, Supporting Information).
   Four tissue cylinders were picked from each patient sample (two from
   normal/benign areas and two from carcinoma regions). In normal/benign
   tissue, the most intense α1‐integrin expression was observed in the
   stromal cells of the prostate and low levels were detected in the basal
   epithelium whereas α2‐integrin was highly expressed by basal epithelial
   cells and moderately by luminal epithelial cells (Figure [89]2g).

Figure 2.

   Figure 2
   [90]Open in a new tab

   Downregulation of ITGA1 or ITGA2 correlates with PCa severity and
   progression. a–f) Boxplots displaying expression levels of ITGA1 a–c)
   or ITGA2 d–f) downregulated during PCa development and progression.
   P‐values were determined by the Kruskal‐Wallis H test for comparing
   three or more groups c and f) or the Mann‐Whitney U test for comparing
   two groups a,b and d,e). g) Representative images of α1‐ and
   α2‐integrin expression and localization in normal and PCa tissue. Scale
   bar = 100 µm. h,i) Kaplan Meier plots indicate increased biochemical
   recurrence h), and metastatic i) risks for PCa patients with tumors
   expressing lower ITGA1 levels. j,k) PCa patients with decreased ITGA2
   expression level are associated with increased risks for biochemical
   recurrence j), and metastasis k). PCa patients were stratified into
   lower and higher expression groups by the median value of ITGA1 or
   ITGA2 expression levels. P‐values were assessed by log‐rank test. l–o)
   Decreased ITGA1 expression level correlates with higher tumor stage l),
   Gleason score m), lymph node metastasis n) and PSA levels o). (p‐s)
   ITGA2 downregulation is associated with higher tumor stage p), Gleason
   score q), lymph node metastasis r) and PSA levels s). P values were
   determined by Kruskal‐Wallis H test.

   This analysis confirmed a significant downregulation of α2‐integrins in
   tumor lesions (Figure [91]S3a,c, Supporting Information). Curiously,
   α1‐integrin expression was significantly reduced in the tumor stroma
   (Figure [92]S3a,b, Supporting Information). These results suggest a
   potential functional role of ITGA1/ITGA2 in prostate tumors. Thus, we
   investigated the association of ITGA1/ITGA2 expression with PCa patient
   prognosis by conducting the Kaplan‐Meier analysis. The results showed
   that the patient group with tumors expressing lower levels of
   ITGA1/ITGA2 had increased risk of biochemical relapse and metastasis in
   multiple independent PCa cohorts (Figure [93]2h–k;
   Figure [94]S2g,h,j,k, Supporting Information). Notably, we also
   observed that PCa patients with lower ITGA1/ITGA2 mRNA levels
   experienced shorter time trends for overall survival (Figure [95]S2i,l,
   Supporting Information). Additionally, we analyzed PCa expression data
   with clinical features, and found that ITGA1/ITGA2 mRNA levels were
   significantly lower in tumors with high tumor stage, Gleason score,
   lymph node and elevated PSA levels in multiple distinct patient cohorts
   (Figure [96]2l–s; Figure [97]S2m–q, Supporting Information). Taken
   together, these results demonstrate prognostic values of ITGA1/ITGA2
   loss/del and downregulation upon PCa progression to metastatic stage,
   suggesting that ITGA1/ITGA2 are new biomarkers that could distinguish
   aggressive disease and might possess tumor suppressive functions in PCa
   development.

2.3. Loss of α1‐ and α2‐Integrins Synergistically Promote Prostate Epithelial
Cell Invasiveness

   Given the strong correlation of α1‐ and α2‐integrin depletion and
   downregulation with PCa progression we analyzed the expression levels
   of α1‐ and α2‐integrin levels in benign prostate epithelial cells
   (RWPE1) and the malignant PCa cell lines (DuCaP, PC3, VCaP, DU145 and
   22Rv1) by Western blotting (Figure [98] 3a; Figure [99]S12, Supporting
   Information). As expected, HPV18‐immortalized benign RWPE1 cells
   expressed high levels of both α1‐ and α2‐integrins. In contrast, brain
   metastasis‐derived DuCaP and DU145 and prostate carcinoma cell line
   22Rv1 displayed reduced levels of α1‐ and especially of α2‐integrins.
   Two bone metastasis‐derived cell lines showed reduced levels of either
   α1‐ (PC3) or α2‐integrins (VCaP). In summary, cancer cell lines
   displayed downregulated levels of either one or both collagen‐binding
   integrins.

Figure 3.

   Figure 3
   [100]Open in a new tab

   Loss of α1‐ or α2‐integrins leads to different but synergistic
   phenotypes in prostate epithelial cells. a) Level of α1‐ and
   α2‐integrins in normal (RWPE1) and PCa (DuCaP, PC‐3, VCaP, DU145,
   22Rv1) epithelial cells. The data shows the mean value from three
   independent experiments, analyzed with One‐way ANOVA. b) RWPE1‐WT,
   RWPE1‐α1KO, RWPE1‐α2KO and RWPE1α1α2dKO cell lysates were analyzed for
   the expression levels of α1‐ and α2‐integrins. β‐tubulin was used as a
   loading control. The data shows the mean value from three independent
   experiments, analyzed with One‐way ANOVA. c) RWPE1‐WT, RWPE1‐α1KO,
   RWPE1‐α2KO and RWPE1‐α1α2dKO cells grown on glass coverslips for 2 days
   were imaged using phase contrast microscopy. Scale bar = 10 µm. d)
   Proliferation of the indicated RWPE1 cell lines were analyzed using an
   XTT assay, analyzed with Two‐way ANOVA. The data shows mean ±SD from
   three independent experiments performed in triplicates. e) Migration of
   the different RWPE1 variants was analyzed using the IncuCyteS3 Scratch
   Wound module. f) A plot showing the wound‐closure dynamics of the
   indicated RWPE1 cell lines. The mean ± SD from a representative assay
   with 3 replicates is plotted in the graph, analyzed with Two‐way ANOVA.
   The assay was repeated thrice with similar results. g) Phase contrast
   microscopy images of the indicated RWPE1 variants grown in 3D Matrigel
   for 7 days. Scale bar = 50 µm. h) Quantitation of the 3D morphology
   analysis of RWPE1‐WT, ‐α1KO and α2KO cell lines. RWPE1‐α1α2dKO cells
   formed interconnected multicellular networks and could not be analyzed.
   Cysts were classified into 4 categories: cysts with a central hollow
   lumen, multilumen cysts, cysts with individual cells in the lumen and
   solid cysts with no visible lumen. The data shows the mean ± SD from
   three independent experiments in which at least 100 cysts per sample
   was analyzed. Statistical significance is indicated by asterisks: ∗ = p
   < 0.05; ∗∗ = p < 0.01; ∗∗∗ = p < 0.001.

   To study the potential functional role of these integrins in prostate
   epithelial cells, we took benign RWPE1 cells, expressing high levels of
   both α1‐ and α2‐integrins and generated α1‐integrin knockout (α1KO) and
   α2KO RWPE1 cells using CRISPR/Cas9‐mediated genome editing.^[ [101]^13
   ^] Two independent gRNAs targeting different exons were designed for
   each gene and used as independent biological replicates in the
   subsequent experiments to confirm the specificity of the KO‐phenotype
   and to exclude possible off‐target effects (Table [102]S2, Supporting
   Information). To address the potential redundance or synergistic
   functions of the two collagen receptors we also established a double
   knockout cell line (α1α2dKO). Specific loss of the target proteins was
   confirmed by Western blotting (Figure [103]3b; Figures [104]S4a and
   [105]S12, Supporting Information). In our culture conditions, parental
   RWPE1 cells and RWPE1 cells transduced with a control virus form a
   cobblestone‐like cell monolayer upon confluency (Figure [106]3c).
   Individual RWPE1‐α1KO cells appeared somewhat rounder than control
   cells and while it took longer for RWPE1‐α1KO cultures to reach
   confluency, they eventually formed a monolayer resembling that of the
   controls. RWPE1‐α2KO cells, and especially the RWPE1‐α1α2dKO cells,
   displayed an elongated shape reminiscent of mesenchymal or actively
   migrating cells (Figure [107]3c). Curiously, RWPE1‐α2KO cells reached
   confluency significantly faster than the other cell lines. To study
   cell proliferation in more detail, we performed an XTT assay. As
   expected, RWPE1‐α2KO cells grew significantly faster whereas RWPE1‐α1KO
   cells proliferated with slower kinetics when compared with control or
   α1α2dKO RWPE1 variants (Figure [108]3d).

   Given that both α1‐ and α2‐integrins are collagen receptors, we
   examined their respective functional roles in collagen adhesion. RWPE1
   control, ‐α1KO, ‐α2KO and ‐α1α2dKO cell variants were seeded onto
   collagen I‐coated substrates and cell spreading in each sample was
   monitored. RWPE1 control, ‐α1KO and ‐α2KO cells all efficiently spread
   on collagen whereas the RWPE1‐α1α2dKO cells lacking both α1‐ and
   α2‐integrins failed to adhere and spread even after long (up to 12
   hours) incubation periods (Figure [109]S4b,c, Supporting Information).
   Thus, collagen‐binding and cell spreading can be mediated by either α1‐
   or α2‐integrins but in the absence of both collagen adhesion is
   severely impaired in prostate epithelial cells. Next, we studied the
   migratory properties of the different RWPE1 variants using the IncuCyte
   S3 Scratch wound module. Confluent monolayers were scratched, and wound
   closure was monitored over a period of 30 hours. In line with the
   observed elongated morphology, wounds in the RWPE1‐α2KO and
   RWPE1‐α1α2dKO monolayers recovered significantly faster than in RWPE1
   control monolayers (Figure [110]3e,f). In contrast, RWPE1‐α1KO cells
   showed significantly impaired wound closure capacity
   (Figure [111]3e,f). Since α1α2dKO cells are incapable of spreading on
   collagen these results suggest that wound closure is mediated by cell
   migration on another cell secreted ECM, presumably fibronectin or
   laminin.

2.4. Loss of α1β1‐ and α2β1‐Integrins Leads to Impaired Acinar Morphogenesis
in 3D

   PCa progression is associated with gradual loss of cell polarity
   leading to aberrant glandular morphology and function. Benign RWPE1
   cells form hollow cysts resembling acinar topology in vivo when
   cultured in 3D gels.^[ [112]^6 ^] To assess morphology of control, α1‐,
   α2‐ and α1α2‐depleted RWPE1 cells, the different variants were seeded
   into 3D basement membrane extract (BME) gels and cultured for up to 7
   days. The vast majority of control cells formed hollow cysts with
   smooth basal surface and an apical lumen occasionally containing few
   individual cells inside the lumen (Figure [113]3g,h; Figure [114]S4d,
   Supporting Information). Only half of the RWPE1‐α1KO cells formed a
   central lumen while basal surfaces had a grape‐like appearance possibly
   due to reduced cell‐cell interactions (Figure [115]3g,h). In contrast,
   80% of RWPE1‐α2KO cysts were smooth‐surfaced solid clusters lacking
   visible apical lumens (Figure [116]3g,h). Surprisingly, the 3D
   morphology of RWPE1‐α1α2dKO cells could not be properly quantitated as
   they hardly formed any cysts and instead grew as solid cell clusters
   connected to each other via chains of collectively migrating cells
   (Figure [117]3g). This phenotype suggests that α1/α2‐dual depleted
   RWPE1 cells possess invasive properties in 3D cultures.

2.5. Loss of α1β1‐ and α2β1‐Integrins Promotes Activation of the EMT Pathway
in Prostate Epithelial Cells

   To understand the role of ITGA1/ITGA2 in PCa, we performed
   transcriptome‐wide RNA sequencing (RNA‐seq) analysis of differentially
   expressed genes (DEGs) upon knockout of ITGA1, ITGA2, or ITGA1&ITGA2 in
   RWPE1 cells (Figure [118] 4a). In each experimental group, two
   biological replicates were included and showed high sample correlations
   (Figure [119]S5a–d, Supporting Information). The RNA‐seq analysis
   identified 3073 upregulated and 2575 downregulated DEGs in
   RWPE1‐α1α2dKO, 408 upregulated and 537 downregulated DEGs in RWPE1‐α1KO
   and 2898 upregulated and 2317 downregulated DEGs in RWPE1‐α2KO cells
   when compared with control (DESeq2, FDR < 0.05; Figure [120]4a).

Figure 4.

   Figure 4
   [121]Open in a new tab

   Loss of both α1‐ and α2‐integrins activates EMT‐associated pathways in
   prostate epithelial cells. a) Heatmap representation of differentially
   expressed genes (FDR < 0.05) identified from the RNA‐seq analysis in
   RWPE1‐WT (control) and RWPE‐α1α2dKO (left panel), ‐α1KO (middle panel)
   or ‐α2KO (right panel) cells. P values were evaluated by the Wald test
   and adjusted for multiple comparisons. b) Top‐ranked pathways enriched
   in the upregulated genes form the GSEA analysis in RWPE1‐ α1α2dKO
   cells. Categories were ranked by the Normalized Enrichment Score (NES).
   c) GSEA enrichment plot displaying the EMT pathway enriched in
   upregulated genes in RWPE1‐α1α2dKO cells. d) Validation of the
   metastasis/EMT‐related pathway signatures in additional resources. e)
   Heatmap representation of genes in the EMT pathways measured by RNA‐seq
   in the RWPE1 cells with α1α2‐dKO. f) The EMT score is significantly
   elevated in PCa tumors with copy loss/del of both ITGA1 and ITGA2. g)
   The EMT score is negatively correlated with the expression levels of
   ITGA1 and ITGA2 in PCa. P value was assessed by the two‐sided Pearson's
   product‐moment correlation test. h) The EMT score is downregulated in
   PCa tumors with higher expression levels of ITGA1/ITGA2. The EMT score
   and the ITGA1/ITGA2 expression levels are calculated as the z‐score
   sum. In f and h, P values were assessed by two‐sided Mann–Whitney U
   test. i) Quantitative PCR analysis confirming regulation of the EMT
   pathway genes, analyzed with One‐way ANOVA. The data is presented as
   mean ± SD of triplicate experiments. ∗ = p < 0.05; ∗∗ = p < 0.01; ∗∗∗ =
   p < 0.001, analyzed with One‐way ANOVA. j) The TGFβ signaling pathway
   was significantly enriched in the upregulated genes with α1α2dKO in the
   RWPE1 cells in multiple gene set resources. k) GSEA enrichment plot
   demonstrating the TGFβ signaling pathway enriched in the upregulated
   genes in RWPE1 cells with α1α2dKO.

   To investigate potential functional categories of ITGA1/ITGA2 KO target
   genes, we performed a gene set enrichment analysis (GSEA) that revealed
   the epithelial‐to‐mesenchymal transition (EMT) pathway to be highly
   enriched in α1α2‐dKO upregulated genes in the Hallmark gene sets
   (Figure [122]4b,c). The interferon response pathways and EMT pathway
   were top enriched in the α1‐ or α2‐KO upregulated genes, respectively
   (Figure [123]S5e,f, Supporting Information). Given the strongest
   morphological phenotype and invasive behavior was observed in the
   RWPE1‐α1α2dKO cells, we focused our analysis on these variants. To
   further consolidate these findings, we performed additional enrichment
   analysis in multiple collections of annotated gene sets. The results
   constantly revealed the strong enrichment of genes upregulated upon
   α1α2‐dKO for pathways relevant with cancer metastasis and EMT
   (Figure [124]4d). Further expression profiling analysis of the EMT
   pathway revealed that the majority of genes in the EMT pathway were
   upregulated upon α1α2‐dKO (Figure [125]4e), which is in line with our
   above observations on the association of ITGA1/ITGA2 copy number
   loss/del and downregulation with PCa tumor progression to advanced
   stages and metastasis (Figure [126]1f–l, Figure [127]2;
   Figures [128]S1d–n, S2, and [129]S3a–c, Supporting Information). We
   next sought to evaluate whether the copy loss/del or expression levels
   of ITGA1/ITGA2 directly correlates with PCa progression, and thus
   further performed correlation analysis with the EMT score that was
   reported to correlate with known EMT markers in PCa patient tumors (see
   Experimental Section/Methods). The result showed that the EMT score was
   markedly upregulated in PCa patients with ITGA1/ITGA2 copy loss/del
   (Figure [130]4f). Consistently, ITGA1/ITGA2 expression levels showed
   negative correlation with the EMT score (Figure [131]4g). We further
   stratified PCa patients by the median expression value of ITGA1/ITGA2,
   and found that in line with previous findings, the EMT score was
   elevated in patient group with ITGA1/ITGA2 low expression levels
   (Figure [132]4h). To further validate these findings, we checked the
   expression levels of selected key EMT markers in the different cell
   lines by quantitative PCR. Consistent with the in vitro phenotype,
   RWPE1‐α1α2dKO cells showed significant upregulation of SLUG, TWIST,
   SNAII, CDH2, ITGA5 and ITGAV and downregulation of CDH1 expression
   levels indicative of EMT (Figure [133]4i). Of note, CDH1 was strongly
   downregulated also in RWPE1‐α1KO cells whereas the RWPE1‐α2KO cells
   showed curious downregulation of SNAIL and CDH2 and neither of the
   single KOs displayed activation of the full EMT program relevant genes
   (Figure [134]4i). Western blot analysis in the different cell lines
   showed robust upregulation of vimentin, another key EMT‐marker
   (Figure [135]S5g, Supporting Information). Interestingly, the GSEA
   analysis in the Hallmark gene sets also revealed the TGFβ signaling
   pathway, representing a subset of the EMT pathway, significantly
   enriched in the upregulated genes identified by RNA‐seq of α1α2dKO in
   the RWPE1 cells (Figure [136]4j,k). Taken together, our transcriptome
   profiling analysis highlighted activation of the EMT pathway and the
   associated TGFβ‐signaling pathway as possible cellular transcriptional
   programs that underlie the potential higher tumorigenic potential of
   ITGA1/ITGA2 loss/del in PCa patients and the invasive behavior in
   α1‐/α2‐integrin dual depleted prostate epithelial cells.

2.6. RWPE1‐α1α2dKO Cells have Elevated Levels of TGFβ1 Secretion Leading to
YAP1 Activation

   To directly assess activation of the TGFβ1 pathway in RWPE1‐α1α2dKO
   cells we harvested culture medium from confluent cultures and analyzed
   levels of TGFβ1 from concentrated medium. TGFβ1 is secreted as a latent
   complex that can associate with several ECM proteins, such as
   fibronectin, and that needs to be mechanically or proteolytically
   activated prior to binding to TGFβ‐receptor proteins at the cell
   surface.^[ [137]^14 ^] RWPE1‐α1KO and RWPE1‐α1α2dKO cells both secreted
   higher levels of TGFβ1 but only in the latter it was found in an
   activated form (Figure [138] 5a,b; Figure [139]S6a,b, Supporting
   Information). We subsequently applied two independent approaches to
   experimentally confirm the role of TGFβ1 in invasive migration. First,
   we treated 3D cultures of control RWPE1 cells with recombinant TGFβ1
   and found that this treatment induced similar invasive structures as
   were seen in RWPE1‐α1α2dKO cells (Figure [140]5c). Second,
   RWPE1‐α1α2dKO cells grown in 3D BME gels were treated with TGFβ
   inhibitor, SB431542, that strongly inhibited the branching capacity of
   these cultures (Figure [141]5c).

Figure 5.

   Figure 5
   [142]Open in a new tab

   TGFβ activation drives invasive potential in α1‐ and
   α2‐integrin‐deficient prostate epithelial cells. a) The culture medium
   from RWPE1‐WT and RWPE1‐α1α2dKO cells was harvested and concentrated as
   described in Experimental Section/Methods. Secreted latent TGFβ and
   activated TGFβ were visualized by western blotting. The data is
   representative of three independent experiments with similar results.
   b) Quantitative analysis of the secreted TGFβ. The data is presented as
   mean ± SD of triplicate experiments. ∗ = p < 0.05; ∗∗ = p < 0.01; ∗∗∗ =
   p < 0.001, analyzed with One‐way ANOVA. c) RWPE‐1‐WT cells were grown
   in 3D BME gel with or without 10 ng of TGFβ1 for 7 days while
   RWPE1‐α1α2dKO cells were grown in the presence or absence of 5 µM TGFβ1
   inhibitor (TGFβ1i). Scale bar: 50 µm. d) RWPE1‐WT and RWPE1‐α1α2dKO
   cells were grown on glass coverslips in the absence or presence of
   10 ng TGFβ1 or 5 µM of TGFβ1i, respectively in coverslips for 2 days,
   fixed and stained for nuclei (blue), actin (red) and YAP1 (green).
   Scale bar:10 µm.

   The transcriptional coactivator YAP1 is a common oncogenic factor that
   promotes proliferation and migration in several cancer types including
   PCa.^[ [143]^15 ^] Integrins are central regulators of YAP1 that is
   activated by the Hippo pathway and has been shown to crosstalk with the
   TGFβ pathway.^[ [144]^16 , [145]^17 ^] We therefore analyzed YAP1
   expression and subcellular localization in the different RWPE1
   variants. In RWPE1 cells YAP1 is mainly cytoplasmic, while in α1α2dKO
   cells it was observed in the nucleus (Figure [146]5d;
   Figure [147]S6c,d, Supporting Information). To study if TGFβ activation
   regulates nuclear translocation of YAP1, we treated RWPE1 cells with
   TGFβ. This resulted in robust nuclear accumulation of YAP1
   (Figure [148]5d). In contrast, inhibition of TGFβ signaling prevented
   nuclear targeting of YAP1 in α1α2dKO RWPE1 cells (Figure [149]5d).
   Thus, TGFβ activation was shown to drive nuclear translocation of YAP1.
   To investigate the functional role of YAP1, we generated lentiviral
   shRNA‐mediated knockdown of YAP1. Depletion of YAP1 in RWPE1
   (Figure [150]S7a,b, Supporting Information) and RWPE1 cells with
   α1α2dKO (Figure [151]S7c,d, Supporting Information) was confirmed at
   both mRNA and protein levels by RT‐qPCR and Western blotting,
   respectively. Interestingly, YAP1‐KD in RWPE1 and PC3 cells led to
   reduced expression levels of ITGA1 and ITGA2 (Figure [152]S7e,f,
   Supporting Information). We next investigated the correlation between
   ITGA1/ITGA2 and YAP1 expression in the human prostate tumors and
   observed significant positive correlation in multiple independent PCa
   cohorts (Figure [153]S7g, Supporting Information). These findings
   suggest that YAP1 might contribute to the regulation of ITGA1/ITGA2
   transcription. Given that YAP1 has been reported to regulate
   transcription by interacting with TEAD‐family members,^[ [154]^18 ^] we
   explored a large collection of genome‐wide chromatin
   immunoprecipitation sequencing (ChIP‐seq) data derived from PCa and
   other tissues including breast and lung to study the regulatory
   mechanisms at the ITGA1/ITGA2 loci. We found an enrichment of chromatin
   binding of TEAD1 and YAP1 transcription factors in the proximity of
   ITGA1 and ITGA2 promoter regions (Figure [155]S7h, Supporting
   Information). To study whether YAP1 is required for the activation of
   TGFβ signaling in α1α2dKO cells, we generated YAP1‐depleted α1α2dKO
   RWPE1 cells (α1α2dKO‐YAP1KD). While α1α2dKO‐YAP1KD cells showed reduced
   YAP1 levels, no significant effect was seen on their ability for
   secreting and activating TGFβ1 (Figure [156]S7i, Supporting
   Information). However, downregulation of YAP1 indeed led to reduced
   proliferation of α1α2dKO cells, suggesting that YAP1 activation
   contributes to the proliferative capacity of α1‐ and α2‐integrin
   depleted PCa cells (Figure [157]S7j, Supporting Information).

2.7. α1α2KO Cells have Enhanced Metastatic Capacity In Vivo

   The parental RWPE1 cells are benign and do not form tumors in vivo.^[
   [158]^6 ^] Given the apparent invasive properties of α1‐ and
   α2‐depleted RWPE1 cells in vitro, we next assessed the metastatic
   capacity of control, α1KO, α2KO and α1α2dKO RWPE1 cells using the
   tail‐vein injection model in immunocompromised SCID mice. All the cell
   lines were transduced to express luciferase and GFP to allow in vivo
   visualization. After 4 weeks the mice were sacrificed, and the lungs
   were imaged using an in vivo imaging system (IVIS). Significant tumor
   foci were seen only for the α1α2dKO‐injected group and weak signals
   were observed in some of the lungs in the α2KO group (Figure [159] 6a).
   Interestingly, detailed histological analysis of the lungs revealed
   several micrometastases in all but one mouse of the α1α2dKO group and
   in two α2KO samples. A single very small lesion was found in one of the
   α1KO samples (Figure [160]6b). As expected, no lesions were found in
   the RWPE1 control samples. The sizes of micrometastases in α1α2dKO
   samples were significantly larger than those found in α2KO cells
   (Figure [161]6c). Given that the capacity of tumor cells to survive in
   circulation is crucial for their metastatic capacity, we collected the
   blood from each mouse and quantitated the number of circulating tumor
   cells (CTCs). Blood could not be collected from one mouse injected with
   α1α2dKO cells. This analysis revealed that significantly more CTCs were
   detected in α1α2dKO and α2KO groups (Figure [162]6d). Collectively,
   these results provide supporting evidence that the benign prostatic
   epithelial cells with co‐deletion of α1‐ and α2‐integrins showed
   enhanced metastatic potential in vivo.

Figure 6.

   Figure 6
   [163]Open in a new tab

   Deletion of α1‐ and α2‐integrins in benign RWPE1 prostate epithelial
   cell line promotes formation of lung micro‐metastases. a) 1 × 10^5
   RWPE1‐WT, ‐α1KO, ‐α2KO and ‐α1α2dKO cells expressing GFP and luciferase
   were injected into the tail veins of 6 immunodeficient SCID‐mice per
   group. After four weeks, mice were sacrificed and the luciferase
   signals originating from cancer cells were measured using IVIS imaging
   system. b) HE‐stained lungs sections were analyzed for presence of
   metastatic lesions. Tumor areas are highlighted by a dotted line. Scale
   bar is 50 µm. c) The area covered by individual lung metastatic lesions
   were determined for each cell variant from HE‐stained lung sections as
   described in methods. The data shows mean ± SD. d) FACS‐analysis of the
   number of circulating GFP‐positive RWPE‐1 cell variants recovered from
   the blood of sacrificed SCID‐mice. For RWPE1‐α1α2dKO cells, blood
   samples were obtained from only 4 mice. Data shows mean ± SD. ∗ = p <
   0.05; ∗∗ = p < 0.01; ∗∗∗ = p < 0.001, analyzed with One‐way ANOVA.

2.8. TEAD1 Transcriptionally Correlates with ITGA1 and ITGA2 and Directly
Regulates their Expression via Chromatin Binding

   Next, we studied the mechanisms of transcriptional regulation of ITGA1
   and ITGA2 expression in PCa. We performed a genome‐wide co‐expression
   analysis of ITGA1/ITGA2 in multiple PCa datasets to prioritize
   potential transcription factors that regulate ITGA1/ITGA2.
   Interestingly, TEAD1 emerged as the top candidate gene that was
   co‐expressed with ITGA1/ITGA2 in the TCGA cohort consisting of 498
   localized PCa tumor samples (Figure [164] 7a–d). Consistently, this
   finding was verified in another PCa cohort comprised of 150 tumor
   samples (Figure [165]S8a–d, Supporting Information). In contrast, we
   observed that the expression levels of the other members of the TEAD
   gene family, TEAD2‐4, were not correlated with either ITGA1 or ITGA2
   (Figure [166]7a,b; Figure [167]S8a,b, Supporting Information). This
   finding suggested that TEAD1 could be a key transcriptional driver of
   ITGA1 and ITGA2.

Figure 7.

   Figure 7
   [168]Open in a new tab

   TEAD1 regulates expression of ITGA1 and ITGA2. a,b) A genome‐wide
   co‐expression analysis identifies strong positive expression
   correlation of ITGA1a) or ITGA2b) with TEAD1, but not with TEAD2, TEAD3
   or TEAD4 in the TCGA cohort. The X‐axis demonstrates Pearson
   coefficient while Y‐axis represents –log10 (P value). c,d) Expression
   correlation analysis revealed significant positive correlation between
   TEAD1 and ITGA1 c) or ITGA2 d) in the TCGA data set. The color bars on
   the right side of figures indicate the expression level of ITGA2 and
   ITGA1, respectively. In a‐d, P‐values were assessed by the Pearson's
   product‐moment correlation test. I Genome browser representation of
   ChIP‐seq signals of transcription factor TEAD1 enriched in the
   promoters and surrounding regulatory regions of ITGA1 and ITGA2 in PC3
   cells. f) ChIP‐qPCR validation of TEAD1 at the three binding sites in
   PC3 cells, analyzed with One‐way ANOVA. g) Knockdown of TEAD1 in the
   RWPE1 cell line downregulates expression of ITGA1 and ITGA2, analyzed
   with One‐way ANOVA. The data is presented as mean ± SD of triplicate
   experiments each with duplicates. h) Western blotting of α1‐ and
   α2‐integrins and TEAD1 in the indicated RWPE1 and PC3 cell variants.
   The blot is representative of three independent experiments with
   similar results. ∗ = p < 0.05; ∗∗ = p < 0.01; ∗∗∗ = p < 0.001, analyzed
   with One‐way ANOVA.

   To experimentally validate this, we performed TEAD1 ChIP‐seq in the
   PC‐3 cell line as previously described,^[ [169]^19 , [170]^20 ^]
   (Figure [171]S8e, Supporting Information). A follow‐up analysis showed
   top‐enriched DNA‐binding motif of the TEAD family transcription factors
   (Figure [172]S8f, Supporting Information), demonstrating the success
   and reliability of the ChIP‐seq experiment. The subsequent genomic
   annotation analysis of ChIP‐seq peaks illustrated that TEAD1 chromatin
   occupancies were greatly enriched in the gene proximal promoter regions
   compared to the genome background (Figure [173]S8g, Supporting
   Information). Notably, among TEAD1 chromatin binding regions, we
   observed an enrichment of TEAD1 binding sites at the bi‐directional
   promoter of ITGA1 and ITGA2 genes and within the ITGA1 gene body
   (Figure [174]7e). To further validate the binding of TEAD1 at ITGA1 and
   ITGA2 gene regulatory regions, we performed ChIP‐qPCR analysis in PC‐3
   cells as described previously.^[ [175]^21 ^] This analysis confirmed
   TEAD1 chromatin binding at the three regions determined by ChIP‐seq
   (Figure [176]7f). To further evaluate TEAD1‐mediated regulation of
   ITGA1/ITGA2 expression, we established TEAD1‐KO cells of RWPE1 and PC3
   and observed reduction of ITGA1 and ITGA2 mRNA levels in TEAD1‐KO RWPE1
   cells (Figure [177]7g). Furthermore, Western blot analysis confirmed
   robust downregulation of α1‐ and α2‐integrins in RWPE1‐ and
   PC3‐TEAD1‐KO cells (Figure [178]7h; Figure [179]S12, Supporting
   Information). Collectively, these results demonstrate that TEAD1
   directly regulates the expression of ITGA1/ITGA2, and this gene
   expression control might also hold true in the clinical setting due to
   their robust co‐expression rates across multiple independent cohorts of
   PCa.

2.9. TEAD1 Expression Correlates with ITGA1 and ITGA2 Expression and is Lost
during PCa Progression

   Given that ITGA1/ITGA2 copy loss and downregulation correlated with PCa
   severity and TEAD1 regulated ITGA1 and ITGA2, we next studied whether
   TEAD1 genomic alterations and expression levels directly correlate with
   PCa clinical characteristics. We found that TEAD1 genomic loss/del
   correlated with lower expression levels (Figure [180] 8a;
   Figure [181]S9a,b, Supporting Information). Notably, PCa patients with
   TEAD1 copy loss/del showed significantly higher risk for biochemical
   recurrence (Figure [182]8b). In line with the association of TEAD1 copy
   loss/del with PCa clinical severity, we observed that TEAD1 expression
   levels were gradually decreased upon PCa development and progression
   compared to normal prostates (Figure [183]8c,d; Figure [184]S9c–e,
   Supporting Information). Moreover, we also observed that decreased
   TEAD1 expression levels were significantly correlated with PCa
   severity, including advanced tumor stage (Figure [185]8e), higher
   Gleason score (Figure [186]8f), lymph node metastasis (Figure [187]8g)
   and PSA levels (Figure [188]8h). Given that TEAD1 was downregulated
   upon disease progression, we studied whether TEAD1 expression may
   possess prognostic value. We conducted the Kaplan‐Meier survival
   analysis and found that low TEAD1 expression significantly correlated
   with an increased risk of biochemical recurrence (Figure [189]8i,j;
   Figure [190]S9f, Supporting Information) and metastasis
   (Figure [191]8k) in PCa patients. To further validate in silico
   analysis, we utilized our Tissue Micro Array (TMA) samples to
   investigate TEAD1 expression in PCa patient tissues. We found that in
   benign samples TEAD1 was highly expressed in the nucleus of basal
   epithelial cells (Figure [192]S9g, Supporting Information). Some
   nuclear TEAD1 expression was also seen in stromal cells
   (Figures [193]S9g and [194]S3a, d, Supporting Information). In
   contrast, we observed little or no TEAD1 staining in cancer tissues
   (Figure [195]8l; Figure [196]S3a, d, Supporting Information). In line
   with these observations, western blot analysis revealed high levels of
   TEAD1 in benign RWPE1 cells whereas some malignant PCa cell lines
   exhibited lower TEAD1 levels (Figure [197]8l; Figure [198]S12,
   Supporting Information). Curiously, TEAD1 levels in the different PCa
   cell lines appeared to correlate with α2‐integrin levels
   (Figure [199]3a). Given the strong correlation of TEAD1 and ITGA1/ITGA2
   expression and the observed regulatory circuit of TEAD1‐ITGA1/ITGA2, we
   next examined whether the RWPE1‐TEAD1‐KO cells display similar
   transformed phenotypes observed in RWPE1‐α1α2dKO cells. While not as
   striking as seen in α1α2dKO cells, confluent RWPE1‐TEAD1‐KO cells had
   slightly more elongated shape (Figure [200]8m). Like α1α2‐dKO cells,
   TEAD1‐KOs showed similar proliferative capacity when compared with the
   controls (Figure [201]8n) but migrated significantly faster than the
   controls in the scratch wound assay (Figure [202]8o). Collectively,
   these results indicate that dampened regulatory circuit of TEAD1 and
   ITGA1/ITGA2 could contribute to PCa progression.

Figure 8.

   Figure 8
   [203]Open in a new tab

   TEAD1 is highly co‐expressed with ITGA1 and ITGA2 and is downregulated
   during PCa development and progression. a) TEAD1 expression levels are
   downregulated upon copy loss/del in PCa. The P value was examined by
   the Mann‐Whitney U test. b) PCa patients with TEAD1 copy number
   loss/del show higher risks for biochemical recurrence. P‐value was
   assessed by the log‐rank test. c,d) TEAD1 expression levels are
   significantly decreased upon PCa tumor progression to metastasis. P
   values were determined by Kruskal‐Wallis H test. e–h) TEAD1
   downregulation correlates with PCa tumor progression to high tumor
   stage e), Gleason score f), lymph node metastasis g) and PSA levels h).
   P values were determined by Kruskal‐Wallis H test or Mann‐Whitney U
   test g). i–k) Lower TEAD1 expression levels in PCa patients are
   associated with higher risks for biochemical recurrence i,j) and
   metastasis k). P‐values were assessed by the log‐rank test. l) Western
   blot analysis of the TEAD1 expression in a benign (RWPE‐1) and
   malignant (DU145, PC3, LnCap, VCap) epithelial prostate cell lines. The
   blot is representative of three independent experiments with similar
   results, analyzed with One‐way ANOVA. m) RWPE1‐WT and RWPE1‐TEAD1KO
   cells grown on glass coverslips for 2 days were imaged using phase
   contrast microscopy. Scale bar = 10 µm. n) Cell proliferation analysis
   of RWPE1 and RWPE1‐TEAD1KO cells was done using XTT assay. The data
   shows mean ±SD from three independent experiments performed in
   triplicates, analyzed with Two‐way ANOVA. o) Cell migration analysis
   using the IncuCyte S3 scratch wound module. The plot shows the mean ±
   SD from a representative assay done in triplicate, analyzed with
   Two‐way ANOVA. The assay was done three times with similar results. ∗ =
   p < 0.05; ∗∗ = p < 0.01; ∗∗∗ = p < 0.001.

2.10. Loss of TEAD1 Phenocopies the α1α2KO Cells

   To further understand the mechanisms by which TEAD1 promotes PCa
   progression, we conducted RNA‐seq in PC3 cells treated with either
   negative control sgRNA or two different sgRNAs targeting TEAD1. Two
   biological replicates were included in each group and high correlations
   were determined between the replicates (Figure [204]S10a–c, Supporting
   Information; r = 0.9994 for control, r = 0.99996 for tEAD1 KOa, and r =
   0.9995 for tEAD1 KOb). We then performed differential gene expression
   analysis and identified 549 and 345 genes significantly upregulated and
   downregulated by TEAD1, respectively (Figure [205]S10d, Supporting
   Information). To better understand how these TEAD1 deregulated genes
   may impact biological processes underlying specific phenotypes, we
   carried out GSEA analysis in five distinct pathway gene sets, including
   Hallmark, Chemical and genetic perturbations (CGP), Reactome,
   WikiPathways (WP), and BioCarta from the MSigDB database. Intriguingly,
   multiple cancer‐promoting pathways, including TGFβ signaling, EMT, and
   metastasis, were repeatedly found highly enriched in upregulated genes
   upon TEAD KO (Figure [206] 9a; Figure [207]S10e, Supporting
   Information). GSEA enrichment plots further indicated that these
   functional pathways were highly enriched in upregulated genes targeted
   by TEAD1 in PC3 cells (Figure [208]9b,c). To observe how TEAD1 affects
   normal prostate cells, we conducted similar RNA‐seq with TEAD1 KO in
   RWPE1 cells. High sample correlations were obtained between two
   biological replicates (Figure [209]S10f,g, Supporting Information).
   Follow‐up differential gene expression analysis identified 1368 and 929
   upregulated and downregulated genes, respectively (Figure [210]S10h,
   Supporting Information). Remarkably, TGFβ signaling, required for tumor
   invasiveness and metastasis, was repetitively found enriched in five
   different pathway gene sets (Figure [211]9d; Figure [212]S10i,
   Supporting Information). GSEA plots indicated upregulation of TGFβ
   signaling pathway in TEAD1‐KO RWPE1 cells (Figure [213]9e,f). Vimentin
   was upregulated in RWPE1‐TEAD1‐KO cells (Figure [214]S5g, Supporting
   Information). These results further support a functional role for TEAD1
   loss in PCa tumorigenesis. To directly study the activation of TGFβ1
   pathway, 3D cultures of control or TEAD1KO RWPE1 cells were treated
   with or without addition of recombinant TGFβ1. Intriguingly, the 3D
   BME‐grown RWPE1‐TEAD1‐KO cells displayed branching morphogenesis in the
   absence of TGFβ, suggesting an invasive phenotype (Figure [215]9g). The
   branching of TEAD1‐KO RWPE1 cells was shown to be TGF‐β1‐dependent as
   TGF‐β1i completely blocked invasive migration in 3D (Figure [216]9h).
   TEAD1‐KO RWPE1 cells also displayed autocrine secretion and activation
   of TGFβ1 (Figure [217]9i). The tumorigenicity of RWPE1‐KO cells was
   analyzed using the tail vein injection model as described for the
   integrin mutants above. RWPE1‐TEAD1‐KO cells formed micrometastases in
   all the injected mice and the size of these lesions was similar as was
   seen for α1α2dKO lesions (Figure [218]9j,k). Moreover, CTCs were
   recovered from the blood of every mouse injected with TEAD1‐KO RWPE1
   cells (Figure [219]9l).

Figure 9.

   Figure 9
   [220]Open in a new tab

   Loss of TEAD1 phenocopies the dual loss of α1‐ and α2‐integrins in
   vitro and in vivo. a) Multiple functional categories, including TGFβ
   signaling, EMT, and metastasis derived from five different pathway gene
   sets from MSigDB database, were repeatedly upregulated upon TEAD1 KO
   profiled by RNA‐seq in PC3 cells. b,c) GSEA plots demonstrating
   pathways relevant to TGFβ signaling, EMT, and metastasis enriched in
   upregulated genes upon TEAD KO in PC3 cells. d) TGFβ signaling were
   commonly found upregulated from five distinct pathway gene sets from
   MSigDB database upon TEAD1 KO profiled by RNA‐seq in RWPE1 cells. PID:
   the Pathway Interaction Database. e,f) GSEA enrichment plots display
   enrichment of TGFβ signaling pathway among upregulated genes following
   TEAD1 KO in RWPE1 cells. g) Phase contrast microscopy images of
   RWPE1‐WT and RWPE1‐TEAD1KO cells grown in 3D BME gels for 7 days. The
   scale bar is 50 µm. h) RWPE1‐ TEAD1KO cells were grown in the presence
   or absence of 5 µM TGFβ1 inhibitor (TGFβ1i) as described in (d). Scale
   bar is 50 µm. i) Culture medium was harvested and concentrated from
   RWPE1‐WT, RWPE1‐α1α2dKO and RWPE1‐TEAD1KO cells. The blot is
   representative of three independent experiments with similar results,
   analyzed with One‐way ANOVA. j) Luciferase expressing RWPE1‐WT and
   ‐TEAD1KO cells were imaged using the IVIS imaging system. Tumor areas
   were highlighted by dotted lines. Scale bar: 50 µm. k) The area covered
   by individual lung metastatic lesions was determined for each cell
   variant from HE‐stained lung sections as described in methods. The data
   shows mean ± SD and was analyzed with One‐way ANOVA l) FACS analysis of
   the circulating WT and TEAD1‐KO RWPE1 cells collected from the blood of
   sacrificed SCID mice. Total GFP‐positive cell count from the sample of
   each mouse and the mean for each group is shown in the plot, analyzed
   with One‐way ANOVA. Data shows mean ± SD. ∗ = p < 0.05; ∗∗ = p < 0.01;
   ∗∗∗ = p < 0.001.

2.11. Synergistic Effects of ITGA1, ITGA2 and TEAD1 Downregulation Promotes
PCa Progression and Severity

   Given downregulation of ITGA1, ITGA2 or TEAD1 apparently associated
   with poor PCa prognosis (Figure [221]2h–k, Figure [222]8i–k;
   Figures [223]S2g–l and [224]S9f, Supporting Information), we sought to
   explore whether ITGA1, ITGA2 or TEAD1 expression levels may hold
   predictive value for stratifying PCa patients with low‐ and high‐risks.
   We hence stratified several cohorts of PCa patients based on Gleason
   scores and investigated the potential correlation between gene
   expression and patient prognosis. The result suggested an explicit
   predictive value of ITGA1, ITGA2 or TEAD1 mRNA levels for patient
   prognosis in PCa group with a Gleason score of 7 (intermediate risk,
   Figure [225] 10a–f), but not for the low‐risk cases with Gleason score
   ≤6 (Figure [226]S11a–f, Supporting Information) nor high‐risk group
   with Gleason score ≥8 (Figure [227]S11g–l, Supporting Information).
   These results indicate ITGA1, ITGA2 or TEAD1 as a potential independent
   prognostic marker in distinguishing PCa patients who are classified
   with the intermediate risks, representing the most difficult patient
   group in clinic to avoid the overtreatment.

Figure 10.

   Figure 10
   [228]Open in a new tab

   The prognostic value of ITGA1, ITGA2 and TEAD1 for PCa patient risk
   stratification and their synergistic impact on PCa in the clinical
   settings a–c) Lower expression levels of ITGA1 exhibit predictive
   values for overall survival (a), biochemical recurrence b), and
   metastasis c) in PCa patient group with an intermediate risk (Gleason
   Score 7). d) PCa tumors with lower expression levels of ITGA2 are
   associated with decreased biochemical recurrence‐free survival in
   patient group with Gleason score 7. e,f) Lower expression levels of
   TEAD1 holds predictive values for increased risks of biochemical
   recurrence e) and metastasis f) in PCa patients with an intermediate
   risk (Gleason score 7). g,h) Forest plots demonstrating triple‐low
   expression of ITGA1, ITGA2 and TEAD1 with high hazard ratio for the
   overall survival g) and biochemical recurrence‐free survival h) of PCa
   patients. i–k) PCa patients with triple‐low expression of ITGA1, ITGA2
   and TEAD1 are associated with shorter overall survival i), increased
   risks of biochemical recurrence j) and metastasis k). In a‐k, P‐values
   were assessed by log‐rank tests. l) Bone metastasis ratio for patients
   stratified based on high or low α1‐integrin, α2‐integrin and TEAD1
   expression status. m–p) PCa patients with triple‐low expression of
   ITGA1, ITGA2 and TEAD1 are associated with advanced tumors including
   Gleason Score m), PSA levels n), tumor stage o), and lymph node
   metastasis p). P values were evaluated by two‐sided Fisher's exact
   test.

   Although we found that decreased expression of ITGA1, ITGA2, or TEAD1
   correlated with PCa severity, it remains unclear whether loss of these
   three genes have synergistic effects on tumor severity compared with
   loss of either individual gene. Therefore, we calculated the expression
   sum of the three genes in pairwise or triple‐wise manner and examined
   the impact on patient prognosis. Intriguingly, the results showed that
   triple low expression of the three genes demonstrated highest hazard
   ratio for the overall survival (Figure [229]10g) and biochemical
   recurrence (Figure [230]10h) of PCa patients. To solidify our findings,
   we stratified PCa patients with triple low or ‐high expression levels
   of ITGA1, ITGA2 and TEAD1 and conducted the Kaplan–Meier analysis in
   multiple PCa cohorts. We found that patients with simultaneous triple
   low expression of ITGA1, ITGA2 and TEAD1 demonstrated significantly
   shorter overall survival (Figure [231]10i), higher risk of biochemical
   recurrence (Figure [232]10j; Figure [233]S11m, Supporting Information),
   and metastasis (Figure [234]10k). Moreover, upon stratification of our
   inhouse TMA patients with expression data of all the three proteins
   combined, we found that patients with triple negative/low expression
   have an increased risk to develop bone metastasis when compared with
   patients with a loss of expression of only one or two of the markers.
   The biggest difference was found when the triple‐negative/low patient
   groups were compared with triple‐positive patient group
   (Figure [235]10l). To further explore the synergistic effects of the
   loss of the three genes in clinical settings, we examined the
   correlation with several clinical features, including Gleason score,
   PSA, tumor stage, and lymph nodes, in PCa cases. The results constantly
   showed that PCa patients with triple low expression of ITGA1, ITGA2 and
   TEAD1 demonstrated higher Gleason score (Figure [236]10m), PSA levels
   (Figure [237]10n; Figure [238]S11n, Supporting Information), advanced
   tumor stages (Figure [239]10o) and lymph node metastasis
   (Figure [240]10p). Taken together, these findings suggested that
   identification of the ITGA1/ITGA2/TEAD1‐triple negative PCa cases may
   have potential in detecting aggressive forms, especially in the most
   problematic Gleason score 7 PCa cases with uncertain PCa clinical
   consequences. Despite the usefulness of PSA screening of potential PCa
   cases, it alone is not a reliable diagnostic tool.^[ [241]^22 ^] Our
   data showed that PCa patient group with triple low expression of
   ITGA1/ITGA2/TEAD1 was markedly associated with poor prognosis,
   increased disease severity and PSA levels, suggesting the
   ITGA1/ITGA2/TEAD1‐axis as a potential biomarker for early detection of
   PCa. Taken together, ITGA1/ITGA2/TEAD1 expression level in the PCa
   specimens could be considered as novel additional biomarkers helping to
   make more accurate treatment decisions in the clinic for PCa patients
   with high PSA levels and intermediate Gleason score 7.

3. Discussion

   PCa progression involves drastic changes in cell‐ECM contacts. A key
   feature observed in PCa is the loss of hemidesmosomes mediated by
   α6β4‐integrins.^[ [242]^6 , [243]^23 , [244]^24 , [245]^25 ^] Our
   meta‐analysis performed here revealed a general downregulation of the
   integrin signaling pathway. While it is thought that invasive PCa cells
   need to interact with the stromal ECM rich in collagen, we found that
   the α‐subunits of the main collagen‐binding integrins, α1 and α2, were
   also strongly downregulated. Importantly, we demonstrated that loss of
   both α1‐ and α2‐integrins not only correlated with PCa progression but
   also contributed to tumorigenesis by inducing autocrine secretion and
   activation of TGFβ, thereby leading to EMT. ITGA1 and ITGA2 are
   neighboring genes in the integrin collagen receptor locus on human
   chromosome 5q11.2. Certain single nucleotide polymorphisms (SNPs)
   within this region have been found to be associated with increased PCa
   risk.^[ [246]^26 ^] We report here that deletion of this locus is a
   relatively common (>2%) event in prostate, ovarian and esophageal
   cancers and deletion and loss as well as downregulation of both ITGA1
   and ITGA2 correlate with development of an aggressive form of PCa.

   Here we identified TEAD1/Tef‐1, encoded by TEAD1 gene, as the key
   regulator of ITGA1/ITGA2 transcription. Out of the four different TEAD
   family members, only TEAD1 expression was strongly correlated with
   ITGA1 and ITGA2 expression. All the four TEAD proteins have been
   implicated in coordination of especially the Hippo signaling pathway,
   but TEADs regulate also Wnt‐, TGFβ‐ and epidermal growth factor
   receptor (EGFR) signaling pathways. Surprisingly, while TEAD
   overexpression has frequently been linked to tumorigenesis, we observed
   here that TEAD1 downregulation or deletion was strongly associated with
   aggressive PCa. Our data suggest that deletion or downregulation of
   TEAD1 contributes to PCa progression by abrogating α1‐ and α2‐integrin
   expression. In contrast to our data, a previous study reported that
   upregulation of TEAD1 correlated with worse PCa prognosis.^[ [247]^27
   ^] The reasons for these discrepant results are unclear but the
   specificity of the TEAD1/Tef‐1 antibodies used for immunohistology
   could perhaps be one contributing factor. In addition to histological
   studies, our current study included analyses of TEAD1 mRNA levels and
   genomic deletion or loss of TEAD1 with all the data supporting a tumor
   suppressive function for TEAD1 in PCa. Curiously, in agreement with our
   data, Knight et al. also reported aberrant 3D morphogenesis in
   TEAD1‐depleted RWPE1 cells. Thus, it seems clear that the role of TEADs
   in PCa needs further scrutiny. It is also noteworthy, that despite
   early promise, recent studies addressing the pharmacological potential
   of targeting the TEAD/YAP1‐axis have raised concerns related to
   activation of alternative oncogenic pathways as seen for TEAD1‐KO cells
   in our study.^[ [248]^28 ^]

   While TEAD1 loss mediates some of its tumorigenic effects by
   influencing α1‐ and α2‐integrin levels there were also apparent
   differences in the mechanisms. First, we found that α1α2dKO led to
   TGFβ‐driven EMT and nuclear translocation of YAP1 to promote
   proliferation. In contrast, although TGFβ was implicated as a driver of
   the invasive phenotype of TEAD1‐KO cells, TEAD1‐KO cells failed to
   activate YAP1, which agrees with earlier studies demonstrating
   TEAD‐dependent nuclear translocation of YAP1.^[ [249]^29 , [250]^30 ,
   [251]^31 ^] Importantly, despite their avid DNA‐binding capacity, TEAD
   family members have limited transcriptional activity by themselves.
   Transcriptional activation depends on their ability to recruit
   cofactors, namely YAP1 and its paralog Taz.^[ [252]^18 ^] Although the
   nuclear YAP1 translocation could be mediated by the other
   TEAD/Tef‐family proteins in TEAD1‐KO cells we did not find evidence
   supporting this in our model. Second, the transcriptomic analysis of
   RWPE1‐α1α2‐dKO and ‐TEAD1‐KO cells revealed significant differences
   with α1α2‐dKO cells displaying robust activation of the EMT signaling
   pathway while TEAD1‐KO cells showed activation of myc‐, TNFα and
   interferon pathways (data not shown). Given the similarity between the
   different TEAD proteins, depletion of TEAD1 might promote aberrant
   activation of the others potentially contributing to PCa oncogenesis.
   Indeed, TEAD2/Tef‐4 has been shown to promote TGFβ‐driven EMT.^[
   [253]^29 ^]

   TGFβ signaling is thought to play a dualistic role in solid tumors.
   While it initially limits cancer growth by inhibiting cell cycle
   progression and inducing apoptosis at later stages of tumorigenesis
   TGFβ signaling promotes proliferation, invasion and drug resistance.^[
   [254]^14 ^] Uncoupling of the proapoptotic signaling arm of TGFβ
   cascade in pretumorigenic cells is thought to convert TGFβ from a tumor
   suppressor into an oncogene. Moreover, TGFβ is a strong immune
   suppressant. Thus, autocrine TGFβ signaling has potentially strong
   protumorigenic effects. TGFβ is secreted in an inactive latent form
   that can be stored in the ECM within the tumor microenvironment.
   Activation of TGFβ is a complex process that can occur via multiple
   mechanisms many of which involve integrin functions.^[ [255]^14 ^]
   Interestingly, we found that cells lacking α1‐integrin activated TGFβ
   production and secretion, but it remained mostly in inactive latent
   form. However, in α1α2‐dKO cells lacking both collagen receptors,
   upregulated TGFβ was efficiently secreted and activated in autocrine
   manner. One potential mechanism underlying TGFβ activation could be the
   observed upregulation of αV‐integrins in α1α2dKO cells. In particular,
   αVβ6‐ and αVβ8‐integrins are critical activators of latent TGFβ.^[
   [256]^32 , [257]^33 ^] Moreover, fibronectin expression is also
   upregulated in α1α2‐dKO cells. Fibronectin is not only a ligand for
   αV‐class integrins, but it also sequestrates latent TGFβ complex and is
   essential for TGFβ activation.^[ [258]^34 ^] Whether an integrin switch
   from laminin‐(α6β4) and collagen‐binding (α1β1 and α2β1) integrins to
   RGD‐binding integrins (αV‐class integrins and α5β1) is indeed driving
   PCa tumorigenesis merits further studies. Interestingly, a bispecific
   antibody targeting αV‐ and α5β1‐integrins, both of which were
   upregulated in α1α2‐dKO cells, was shown to inhibit PCa
   tumorigenicity.^[ [259]^35 ^] In addition, high expression levels of
   αVβ5‐ and αVβ3 integrins have been detected in metastatic PCa.^[
   [260]^36 , [261]^37 ^]

4. Conclusions

   Taken together, our data identifies downregulation of α1‐ and
   α2‐integrins as a risk factor in PCa. Genomic loss or deletion of
   ITGA1/ITGA2 locus was found to be highly frequent in PCa cases and
   equally prevalent in ovarian and esophageal cancers, suggesting that
   the tumor suppressor functions of α1‐ and α2‐integrins might not be
   limited to PCa. Loss of these collagen receptors triggered autocrine
   TGFβ‐signaling and subsequent EMT leading to increased invasive and
   metastatic potential of α1/α2‐integrin deficient PCa cells in vitro and
   in vivo. α1/α2‐depleted cells also activated YAP1 signaling, that
   contributed to cell proliferation in these cells. Moreover, TEAD1 was
   identified as a key transcriptional regulator of ITGA1 and ITGA2 whose
   downregulation also contributed to PCa progression. The
   TEAD1/ITGA1/ITGA2‐axis can potentially be exploited for both
   diagnostics purposes and for the development of more potent therapies
   targeting malignant forms of PCa.

5. Experimental Section

Cell Culture

   Benign RWPE‐1 prostate epithelial cells (ATCC CRL‐11609) were cultured
   in in K‐SFM + Human recombinant Epidermal Growth Factor (rEGF) and
   Bovine Pituitary Extract (BPE) supplemented with 1% penicillin and
   streptomycin (all from Thermo Fisher Scientific). PCa cell line PC‐3
   (CRl‐1435), VCap (CRL‐2876), DuCap (CVCL_2025), 22Rv1 (CRL‐2505), LNCaP
   (CRL‐1740) and DU145 (HTB‐81) were purchased from ATCC. VCaP and
   293T‐D10 (a kind gift from Dr. Peppi Karppinen, University of Oulu)
   were grown in Dulbecco's Modified Eagle's Medium (DMEM) (11 965 092,
   Thermo Fisher Invitrogen), for DU145 Eagle's Minimum Essential Medium
   (EMEM) (30‐2003, ATCC) was used. LNCaP, VCaP and 22Rv1 were grown in
   Roswell Park Memorial Institute Medium (RPMI 1640) (R8758, Sigma). PC‐3
   were grown in F12‐K (30‐2004, Invitrogen). DMEM, EMEM and RPMI were
   supplied with a final concentration of 10% fetal bovine serum
   (16 000 044, Thermo Fisher) and 1% of penicillin and streptomycin
   (15 140 122, Thermo Fisher). All cell lines were cultured in humidified
   incubators at 37 °C and 5% CO[2].

Antibodies and Reagents

   Primary antibodies were listed in Table [262]S1 (Supporting
   Information). All secondary antibodies, nonspecific sheep, rabbit, and
   mouse IgG isotype controls were purchased from Jackson ImmunoResearch.
   TRITC‐phalloidin and DAPI were from Merck.

Lentivirus‐Mediated Gene Knockout

   Gene knockouts using the lentiviral CRISPR/Cas9‐methodology were
   generated as described previously.^[ [263]^13 , [264]^38 ^] In short,
   two separate target sequences from constitutive early exons were
   designed for each gene (Table [265]S2, Supporting Information). Target
   sequences with no predicted off‐target sites (a minimum of three
   mismatches with any other site in the human genome (GRCh38.p13)) were
   selected based on FASTA similarity search tool (EMBL‐EBI). For ITGA1,
   ITGA2 and TEAD1‐KO1 constructs gRNA oligos with BsmBI (New England
   Biolabs) overhangs were ordered from Sigma–Aldrich and subcloned into
   lentiCRISPRv2‐puro (Addgene plasmid # 52 961, a gift from Feng Zhang)
   or lentiCRISPRv2‐neo (Addgene plasmid #98 292, a gift from Feng Zhang)
   and used for lentivirus preparation. The TEAD1‐KO2 lentiviral construct
   was purchased from the Biomedicum Functional Genomics Unit at the
   Helsinki Institute of Life Science and Biocenter Finland at the
   University of Helsinki. To produce lentiviruses, 70%–80% confluent
   293T‐D10 on CellBind 10 cm Ø tissue culture dishes (Corning) were
   co‐transfected with lentiCRISPR (Addgene plasmid 52 961), pPAX2 and
   VSVg plasmids using Lipofectamine 2000 reagent (Thermo Fisher
   Scientific) in Opti‐MEM (Thermo Fisher Scientific). For infection,
   subconfluent cultures were incubated with virus‐containing supernatants
   in the presence of 4 µg mL^−1 polybrene (Sigma–Aldrich, TR‐1003) for
   24 h, virus was removed, and culture continued for 24 h, as described
   before.^[ [266]^13 ^] Cells were then trypsinized and reseeded into
   medium containing 3 µg mL^−1 of puromycin to select transduced cells.
   Clonal cell lines were established from resistant populations. To
   generate double α1/α2‐KO cells, clonal α1KO cells were transduced with
   α2‐targeting lentiCRISPRv2‐neo vector as described above, expanded and
   selected by different antibiotic selection, puromycin for α1‐targeting
   lentiCRISPR vector and 500 µg mL^−1 of neomycin for α2‐targeting
   lentiCRISPR vector.

Lentivirus‐Mediated Gene Knock‐Down

   The lentiviral pLKO.1‐based shRNA expressing vectors targeting YAP1,
   were purchased from the Biomedicum Functional Genomics Unit at the
   Helsinki Institute of Life Science and Biocenter Finland at the
   University of Helsinki. Second generation lentiviral vectors were
   packaged using 293T‐D10 cells, as described above. Target cells were
   transduced sequentially two times, and the transduced cell population
   was selected using 1 µg mL^−1 puromycin (Sigma) as described
   previously.^[ [267]^39 ^]

3D Cell Culture

   One hundred µls of Matrigel (Corning, #354 230) was layered onto a
   3.5 cm high glass bottom cell culture‐dish (IBIDI, #81 156) and allowed
   to solidify for 30 min at + 37 °C (5% CO[2]). Control and α1KO, α2KO or
   α1α2dKO RWPE1 cells grown to confluency were trypsinized and counted.
   Five thousand cells per sample were resuspended into 100 µl of ice‐cold
   Keratinocyte‐SFM containing 2% of FBS and 2% (v/v) of Matrigel.
   Resuspended cells were seeded onto Matrigel‐coated dishes and incubated
   for 10 min at + 37 °C after which 1 ml of Keratinocyte‐SFM medium
   containing and 2% Matrigel was added. Matrigel‐containing medium was
   subsequently refreshed every two days. One week later, cells were
   fixed, stained using DAPI (Sigma, #D9542‐10) and filamentous actin
   (Sigma, #A5441), and analyzed by using an Olympus FluoView FV1000
   confocal microscope or Olympus Cellsense both equipped with 20x UPLSAPO
   objective (NA = 0.75).

Immunofluorescence Staining and Microscopy

   Cells were seeded on slides (2D culture) or on IBIDI glass bottom
   dishes (3D culture) and were cultured until desired confluency. Cells
   were fixed with 4% PFA in PBS+/+ (PBS with 0.9 mM CaCl[2] and 0.5 mM
   MgCl[2]) for 15 min at room temperature for slide and 30 min for IBIDI.
   Immunofluorescence staining was performed as previously described ^7.
   Confocal images were acquired with the Zeiss LSM 780 laser scanning
   confocal microscope using 40x Plan‐Apochromat objective (NA = 1.4) or
   with Olympus FluoView FV100 confocal microscope using 20x and 40x
   objective (NA = 0.75). Image acquisition software was ZEN (black
   edition, LSM 780; blue edition Cell Observer) and with Olympus FluoView
   viewer software, respectively.

SDS‐PAGE and Western Blotting

   Confluent cultures were washed in PBS‐/‐ (Gibco) and scraped into RIPA
   buffer: 0.5% SDS,150 mM NaCl, 10 mM Tris‐HCl pH 8.0, 1% IGEPAL, 1%
   sodium deoxycholate with 2 mM PMSF (phenylmethylsulfonyl fluoride),
   10 µg mL^−1 aprotinin, and 10 µg mL^−1 leupeptin. Protein concentration
   was calculated using BCA Protein Assay Kit (Pierce). 30 ug of protein
   lysate were separated on 7,5% SDS‐PAGE gels and transferred onto a
   Protran pure 0.2 micron nitrocellulose (Perkin Elmer) for 1 h 30 min in
   20% ethanol, 0.192 M glycine and 0.025 M Tris. The membranes were
   blocked for 1 h in 5% skimmed milk and probed with specific primary
   antibodies (Table [268]S1, Supporting Information) overnight at 4 °C.
   Secondary antibodies conjugated with HRP and Lumi‐Light Western
   Blotting Substrate (Roche) were used to visualize specific protein
   bands. The bands were detected using Fujifilm LAS‐3000 bioimaging or
   AZURE 600 and scientific research imaging equipment (Fuji photo film
   co., LTD).

Measurement of Secreted TGF‐b

   Ten milliliters of K‐SFM with supplements was added onto confluent
   cultures of the different RWPE1 variants on Ø 10 cm tissue culture
   dishes. Forty‐eight hours later conditioned medium (CM) was collected
   and centrifuged at 168xg for 5 min to remove cells and debris. CMs were
   concentrated using Amicon Ultra (Merk Millipore UFC503096) columns
   centrifuged at 101xg for 10 min. The 100x concentrated CM was analyzed
   for TGFβ by Western blotting.

Adhesion Assay

   Control, α1KO, α2KO and α1α2dKO RWPE1 cells were seeded onto 6‐well
   plates (Corning‐ Sigma) coated with 20 µg mL^−1 collagen‐I (Advanced
   BioMatrix, 5153‐A). After 30 min, plates were placed into Incucyte S3
   automated imaging system (Essen Bioscience) and phase contrast image
   were taken for 24 with 3 h intervals using Brightfield 20x objective.
   Posterior analysis of the attached cells was performed by counting the
   cells that were properly attached and the average cell count were
   subjected for statistical analysis.

XTT Assay

   4000 cells were seeded on 96‐wells plate and allowed to grow overnight.
   The Cell Proliferation Kit II XTT (11 465 015 001, Roche) was used
   according to the manufacturer. Cell proliferation was examined at day
   1, 3, 5 and 7 by XTT colorimetric assay (absorbance at 450 nm). At
   least three replicate wells were prepared per condition and the data
   were statistically analyzed with a two‐tailed Student's t test.

Scratch Wound Assay

   Cells were seeded into 96 well plates and allowed to grow to
   confluence. The scratch wound was done using the Incucyte Cell
   Migration Kit (Essen Bioscience 4493) and the wells were washed twice
   with PBS to remove detached cells. Fresh culture medium was added, and
   the wound areas were imaged at 3 h intervals using Incucyte Zoom (Essen
   Bioscience) equipped with a 20x objective. The wound area closure was
   analyzed using the Incucyte Zoom software (Essen Bioscience).

Chromatin Immunoprecipitation (ChIP)

   The cells on tissue culture plates were cross‐linked in final
   concentration of 1% formaldehyde for 10 min at room temperature.
   Glycine was added to a final concentration of 125 mM to stop the
   reaction. Cell pellets were collected and snapped frozen in liquid
   nitrogen. Pellets were then suspended in hypotonic lysis buffer (20 mM
   Tris‐Cl, pH 8.0, with 10 mM KCl, 10% glycerol, 2 mM DTT, and Complete
   protease inhibitor cocktail (Roche)) and incubated for up to 50 min to
   isolate the nuclei. The nuclei were washed twice with cold PBS and
   lysed in SDS lysis buffer (50 mM Tris‐HCl, pH 8.1, with 0.5% SDS, 10 mM
   EDTA, and cOmplete Protease Inhibitor). An average size of 400 bp of
   chromatin was prepared by sonication (Q800R sonicator, Q Sonica).
   Seventy µl of Dynabead protein G (Invitrogen) slurry per each reaction
   was washed twice with blocking buffer (0.5% BSA in IP buffer), followed
   by 10 h incubation with 7 µg of TEAD1/Tef‐1 antibody against the target
   protein or control IgG in 1000 µl of 0.5% BSA in IP buffer (20 mM
   Tris‐HCl, pH8.0, with 2 mM EDTA, 150 mM NaCl, 1%Triton X‐100, and
   Protease inhibitor cocktail). Sonicated chromatin lysate (200‐250 µg)
   was diluted into 1.3 ml of IP buffer and added onto bead/antibody
   complexes followed by incubation at 4°C for at least 12 h. Next, the
   complex was washed once with wash buffer I (20 mM Tris‐HCl, pH 8.0,
   with 2 mM EDTA, 0.1%SDS, 1% Triton X‐100, and 150 mM NaCl) and once
   with buffer II (20 mM Tris‐HCl pH, 8.0, with 2 mM EDTA, 0.1% SDS, 1%
   Triton X‐100, and 500 mM NaCl), followed by two washes each with buffer
   III (10 mM Tris‐HCl, pH 8.0, with 1 mM EDTA, 250 mM LiCl, 1%
   Deoxycholate, and 1% NP‐40) and buffer IV (10 mM TrisHCl, pH 8.0, and
   1 mM EDTA), respectively. A hundred ul of extraction buffer (10 mM
   TrisHCl, pH 8.0, 1 mM EDTA, and 1% SDS) was added to extract the
   DNA‐protein complexes from the beads. Proteinase K (5 µl from
   20 mg mL^−1 stock) and NaCl final 0.3 M were added into the complexes
   incubating overnight at 65 °C to reverse the cross‐links of protein‐DNA
   interactions. Finally, DNA was purified with MinElute PCR Purification
   Kit (Qiagen) and the target DNA fragments were analyzed by qPCR
   (Table [269]S5, Supporting Information).

Quantitative RT‐PCR

   RNA was isolated from cultured cell lines using RNeasy Mini Kit
   (QIAGEN), while DNA in these samples were removed by RNase‐Free DNase
   (QIAGEN). The RevertAid reverse transcriptase Kit (Thermo Scientific)
   was used to synthesize cDNA from 2 ug RNA. The SYBR Select Master Mix
   (Applied Biosystems) was used in the Quantitative RT‐PCR reactions.
   High specificity primers were selected for each target (Table [270]S4,
   Supporting Information). Primer sequences are listed in Table [271]S3
   (Supporting Information). For the analysis of mRNA levels, each gene
   was analyzed at least in triplicate and the data was normalized against
   an endogenous GAPDH control. For ChIP‐qPCR, all target primers had
   three technical replicates and the data were normalized to the control
   regions, then the relative enrichment of the target antibodies at
   target DNA fragment were determined by compared with the background
   (IgG control).

Immunohistochemical Staining

   The tissue slides were deparaffinized by incubation for 1 h at 55 °C.
   To rehydrate, they were washed twice in xylene preceded 100% ethanol,
   94% ethanol and 70% ethanol washes (5 min in each). Heat mediated
   antigen retrieval was executed by incubating the slides in boiling
   citric acid buffer (pH 6.0) for 10 min, then washed in PBS‐/‐,
   incubated with 3% H[2]O[2] for 10 min and then blocked with 5% BSA in
   PBS‐/‐ for 1 h. After overnight incubation with primary antibodies at
   4 °C, the slides were washed in PBS‐/‐ and probed for 1 h with HRP
   conjugated secondary antibody. The positive antibody signal was
   revealed using DAB Substrate Kit (Abcam, ab64238). Slides were stained
   for 2 min in Harris hematoxylin solution followed by dehydration steps
   in 70%, 94% and 100% ethanol and two 5 min treatments in xylene and
   then imaged using Zeiss Axio Imager.M2 microscope.

Cancer metastasis In Vivo Model

   All animal experiments were approved by the Finnish National Animal
   Experiment Board (permissions ESAVI/3901/2021) and conducted in the
   Laboratory Animal Centre of the University of Oulu according to the
   principles of 3R (reduction, refinement, replacement). Intravenous tail
   injections were performed on 7 weeks old male IcrTac:ICR‐Prkdc (SCID)
   mice (Taconic Biosciences A/S) as described in.^[ [272]^6 ^] In short,
   1 × 10^5 RWPE1 cell variants expressing luciferase and GFP were
   injected into the lateral tail vein (WT (n = 6), α1KO (n = 6), α2KO (n
   = 6), α1α2dKO (n = 6) and TEAD1KO (n = 6)). One mouse from each group
   except the TEAD1KO‐group died from unrelated causes during the
   experiment and were omitted from the analysis. Mice were monitored on a
   weekly basis using the in vivo imaging system (IVIS) Spectrum CT in
   vivo bioluminescence imaging system (PerkinElmer). Body weight was
   monitored every 3 days and no significant differences in body weights
   between the groups were observed. For analysis, the mice were injected
   with D‐luciferin (Caliper Lifesciences) at 150 mg k^−1 g
   intraperitoneally (IP) and imaged 4 times at 5 min intervals. Four
   weeks post injection, mice were euthanized, the blood was collected for
   analysis of circulating cancer cells and the lungs were dissected for
   analysis of luminescence and GFP signals using the IVIS system. The
   lungs were divided in two, one was embedded for cryosection (Fisher
   Healthcare Tissue‐Plus O.C.T Compound, 23‐730‐571, Fisher & Paykel
   Healthcare AB, Helsinki, Finland) and one for formalin‐fixed paraffin
   embedded (FFPE) sections. FFPE sections covering the entire lung were
   cut and stained with hematoxylin and eosin (HE). Slides were screened
   and metastatic lesions were captured using Zeiss Axio Imager M2
   microscope with 40x objective. The size of micrometastatic lesions was
   determined based on sections where the individual lesions covered the
   largest area. The areas were measured using ImageJ.

The Recovery of Circulating Cancer Cells

   0,5‐0,6 ml of blood from mice was immediately diluted in RBC lysis
   buffer (0.8% NH4Cl, 0.084% NaHCO3, 0.037% EDTA), mixed for 5 min,
   washed in PBS‐/‐ and the recovered cells were seeded onto 10 cm dishes
   in K‐SFM. After 2 days the cells were harvested by trypsin‐treatment,
   washed in PBS‐/‐ and GFP‐positive cells were sorted using BD FACS Aria
   II (BD Biosciences).

META Analysis and Pathway Enrichment Analysis

   We incorporated a total of nine PCa data sets from the GEO repository
   for conducting the meta‐analysis to identify differentially expressed
   genes among normal prostate tissues, primary and metastatic tumors. The
   nine PCa data sets ([273]GSE3325, [274]GSE3933, [275]GSE6099,
   [276]GSE8511, [277]GSE21034, [278]GSE27616, [279]GSE35988‐[280]GPL6480,
   [281]GSE35988‐[282]GPL6848 and [283]GSE62872,^[ [284]^40 , [285]^41 ,
   [286]^42 , [287]^43 , [288]^44 , [289]^45 , [290]^46 , [291]^47 ^])
   were downloaded by R package “GEOquery” with respective GEO IDs.
   Microarray expression data from each study was first normalized by
   logarithm transformation. Differential gene expression analysis was
   conducted on normalized expression data by R package “limma” .^[
   [292]^48 ^] We applied adjustment to raw P values by using Benjamini &
   Hochberg method for multiple testing. Gene symbols were added with
   respective GEO Platform (GPL) identifiers. Each gene list was further
   filtered by removing probes corresponding to more than one gene, and
   unannotated probes and incomplete data was filtered out. Genes with
   multiple probes were selected based on the lowest P‐value. We used R
   package″ MetaVolcanoR″^[ [293]^49 ^] to incorporate multiple PCa
   cohorts with microarray data measured from different platforms for
   conducting the meta‐analysis analysis. We applied both the vote
   counting approach that identifies differential expressed genes (DEG)
   for each study, and the Combining‐approach that summarizes the fold
   change of a gene in different studies by the mean and calculates the
   gene differential expression P values using the Fisher method. We
   integrated results from both methods, applied a stringent cutoff by
   selecting differentially expressed genes with significant P values, and
   followed the same direction of being up or downregulated in at least 6
   out of 8 data sets. We finally generated a comprehensive set of 1519
   differentially expressed genes with 403 and 1116 up‐ and downregulated
   genes, respectively. We next used PANTHER Pathway,^[ [294]^50 ^] to
   investigate the underlying biological mechanisms of this DEG set in PCa
   and identified the integrin signaling pathway as the top altered
   cellular signaling cascade.

Survival Analysis

   The Kaplan‐Meier survival analysis was conducted to evaluate the impact
   of the copy number loss& deletion or low expression level of ITGA1,
   ITGA2 or TEAD1 on patient prognosis in multiple independent PCa data
   sets. Patients were stratified based on the copy number loss & deletion
   or the median expression level of ITGA1, ITGA2 or TEAD1. For the
   investigation of the synergistic effect of ITGA1, ITGA2 or TEAD1 on
   patient survival, we included PCa patients with consensus triple high
   or low expression levels of ITGA1, ITGA2 and TEAD1. To investigate
   whether ITGA1, ITGA2 or TEAD1 possess prognostic value for PCa patients
   with intermediate risk, we pre‐stratified patients by Gleason score > =
   6, 7 or > = 8 that represents low‐, intermediate‐ and –high risks,
   respectively. Kaplan‐Meier survival analysis were conducted using R
   package “Survival” (v. 3.2.13),^[ [295]^51 , [296]^52 ^] and assessed
   by using log‐rank test. Cox proportional hazards model,^[ [297]^53 ^]
   was applied to calculate the hazard ratio (HR) for assessing the
   relative risk between different patient groups.

RNA‐Sequencing (RNA‐seq) and Differential Gene Expression Analysis

   The RNA‐seq library of double knockout of ITGA1 and ITGA2, single
   knockout of ITGA1, ITGA2, or TEAD1 generated paired‐end raw sequence
   reads of 150 bp. Fastq files were first processed by FastQC for quality
   evaluation. Trimmomatic^[ [298]^54 ^] was subsequently used on raw
   sequence reads for quality control. A second run of FastQC was applied
   on cleaned reads to guarantee high read quality. STAR (v2.7.2a)^[
   [299]^55 ^] was applied to align processed reads to the GRCh38 human
   genome reference by default settings. Uniquely mapped reads were then
   quantitated with parameters “‐s no, ‐I gene_name” using HTSeq‐count,^[
   [300]^56 ^] with gene annotation information in gene transfer format
   (GTF) from Encode. Genes with low expression counts (<2 cumulative read
   count across samples) were excluded before conducting the differential
   expression analysis by DESeq2 (1.26.0). Genes with FDR <0.05 were
   identified as differentially expressed. Data normalization was
   processed by the Variance Stabilizing Transformation (VST) method from
   DESeq2. Heatmaps comparing gene expression levels across samples were
   generated by R package “pheatmap” (1.0.12).

Gene Set Enrichment Analysis (GSEA)

   GSEA,^[ [301]^57 ^] was used to interpret the biological functions
   underlying the altered gene expression profiles by the double knockout
   of ITGA1/ITGA2 or single knockout of ITGA1, ITGA2, or TEAD1 in the
   MSigDB database.^[ [302]^58 ^] We sorted genes in a descending order by
   “stat” statistics from the previously generated differential result
   table. The produced pre‐ranked gene list was then used as input for
   conducting the GSEAPreranked test. Parameters were set as default
   except of following modifications: Enrichment statistic = “weight”, Max
   size (exclude larger sets) = 5000, number of permutations = 1000. The
   GSEA enrichment plots were produced by R packages “clusterProfiler”
   (3.14.3) and “enrichplot” (1.6.1).^[ [303]^59 ^]

The EMT Pathway, EMT Score and Gene Expression Correlation Analysis

   The EMT score was generated in the study carried out by Bayer et al.^[
   [304]^60 ^] The authors reported that the EMT signature correlates with
   known EMT markers and could serve as an indicator. The EMT signature
   was devised with a panel of 76 representative genes, including ANKRD22,
   ANTXR2, AP1M2, AXL, BSPRY, C1ORF116, KDF1, XXYLT1, CARD6, CDH1, CDH3,
   CDS1, CLDN4, CLDN7, CRB3, DSP, ELMO3, ENPP5, EPB41L5, EPHA1, EPN3,
   EPPK1, ERBB3, EVPL, F11R, FN1, FXYD3, GALNT3, GALNT5, ADGRF1, ADGRG1,
   GRHL1, GRHL2, HNMT, PATJ, ITGB6, KLC3, KRT19, KRTCAP3, LIX1L, TSKU,
   MAL2, MAPK13, MMP2, MPP7, MPZL2, TC2N, MUC1, NRP1, PPARG, PRR5, PRSS22,
   PRSS8, RAB25, ESRP1, RBPMS, S100A14, SCNN1A, SERINC2, SH3YL1, SHROOM3,
   SPINT2, SSH3, ST14, STAP2, EPCAM, TACSTD2, TGFBI, TJP3, TMC4, TMEM125,
   TMEM30B, TMEM45B, TNFRSF21, VIM, ZEB1. The expression values of the EMT
   signature were calculated as z‐score sums. The EMT pathway, consisting
   of 200 genes, was retrieved from the Hallmark gene sets from the MSigDB
   database.^[ [305]^57 , [306]^58 ^] The gene co‐expression analysis was
   examined by the Pearson's product‐moment correlation. For the
   genome‐wide co‐expression analysis that identified TEAD1 as one of the
   genes most correlating with ITGA1 and ITGA2, we first investigated the
   expression correlation of genes on a global scale with ITGA1 or ITGA2
   by Pearson correlation test. Gene ranks were represented by the Pearson
   coefficient and log10(P‐value) in X‐ and Y‐axis, respectively.

Chromatin Immunoprecipitation Sequencing (ChIP‐seq)

   The ChIP‐seq library was sequenced, which generated 150 bp long single
   end reads. FastQC was used to check the quality of raw sequence reads.
   Trimmomatic,^[ [307]^54 ^] was then applied for the quality control
   process with following parameters: TruSeq3‐SE.fa:2:30:10
   SLIDINGWINDOW:5:20. Read length less than 10 bp were filtered out.
   Bowtie2^[ [308]^61 ^] was then used to map the processed reads to the
   human genome hg38 with default settings. MACS2^[ [309]^62 ^] was
   applied for calling peaks with default settings. UCSC tools were
   applied to create ChIP‐seq coverage signals and Integrative Genomics
   Viewer (IGV),^[ [310]^63 ^] was used for visualization.

Univariate Analysis

   For the univariate analysis, we investigated the association of the PCa
   patients’ overall survival, biochemical recurrence and metastasis with
   single, pairwise, and triple‐wise combinations of gene expression
   levels of ITGA1, ITGA2 and TEAD1. The z‐score sum of gene expression
   was calculated and PCa patients were then stratified into higher and
   lower groups by the median expression values of the cumulative
   expression levels. The analysis was conducted by using R packages
   “tidyverse”,^[ [311]^64 ^] “tidytidbits”^[ [312]^65 ^] and
   “survivalAnalysis”^[ [313]^66 ^] Statistics were summarized and
   presented in the forest plot.

Meta‐Analysis for Time‐to‐Event Outcomes

   The meta‐analysis for investigating the association between the ITGA1
   or ITGA2 copy number loss/del and patient prognosis across multiple PCa
   cohorts was conducted using the R package “metafor” (v.3.4.0) in the R
   environment v.4.2.0.^[ [314]^67 ^] The pooled HR was calculated by a
   fixed effect model,^[ [315]^68 ^] provided the I^2 statistic was less
   than 30% or the fixed effects P value for the I^2 statistic was greater
   than 0.10, indicating insignificant heterogeneity across studies.^[
   [316]^69 ^]

Statistical Analysis

   The RNA‐seq, microarray or clinical data was acquired from the
   cBioPortal for Cancer Genomics,^[ [317]^70 , [318]^71 ^] NCBI Gene
   Expression Omnibus (GEO) repository ^49, Oncomine,^[ [319]^72 ^] and
   literature. Microarray data for the Meta analysis was uniformly
   normalized by logarithm transformation (log2). Statistical tests
   applied across normal prostate tissues, primary and/or metastatic
   tumors were evaluated by meta‐analysis, Mann‐Whitney U test or
   Kruskal‐Wallis H test depending on the context and the number of
   groups. The correction between gene expression data and clinical
   variables were assessed by Mann‐Whitney U test or Kruskal‐Wallis H
   test. The proportion of PCa patients with primary or metastatic tumors
   presenting the degree of number loss and deletion of ITGA1 or ITGA2 was
   examined by fisher‐exact test. For investigating the synergistic effect
   of ITGA1, ITGA2 and TEAD1, PCa patients were stratified to high and low
   groups by the median expression values of ITGA1, ITGA2 and TEAD1,
   respectively. Patients with consensus high or low expressions of ITGA1,
   ITGA2 and TEAD1 were labeled as triple high or low group, respectively.
   The correlations between the synergic expression of ITGA1, ITGA2 and
   TEAD1 with different clinical variables including PSA, lymph node,
   tumor stage, and Gleason score were investigated by fisher‐exact test.
   For results from microarray‐based expression profiling, gene probes
   with lowest P values were selected. Samples with missing gene
   expression or patient survival data were excluded from analyses. P
   value <0.05 was considered to be statistically significant. For
   boxplots, the interquartile range (IQR) is depicted by the box with the
   median represented by the center line. Whiskers maximally extend to
   1.5 × IQR (with outliers shown). The above‐mentioned statistical
   analyses were carried out by R (v. 4.1.0) and RStudio (v.1.4.1106).
   Sample size from each analysis was mentioned in relevant figure or
   figure legend. Data were expressed as means ± SD of at least three
   independent experiments. Comparative data were analysed with the
   unpaired or paired Student's t‐test or one‐way ANOVA as indicated using
   GraphPad Prism 8 software. The results were considered statistically
   significant when the P value was less than 0.05 (*), 0.01 (**) or 0.001
   (***).

Ethical Approval Statement

   The Oulu PCa cohort samples derived from radical prostatectomy patients
   were obtained under written informed consent with approval of the
   Regional Ethics Committee of the Northern Ostrobothnia Hospital
   District (EETTMK 4/2015) in compliance with the Declaration of
   Helsinki. Mouse cancer models and protocols were approved by the
   National Animal Experiment Board (ESAVI/3901/2021). In all animal work,
   the principle of 3R (reduction, refinement, replacement) was respected.

Conflict of Interest

   The authors declare no conflict of interest.

Author Contributions

   S.C.P and Q.Z. contributed equally to this work. A.M., G.H.W. designed
   and conceptualized the work. S.C.P. performed most of the experiments
   and analyzed the data. Q.Z. carried out and interpreted all
   bioinformatics analysis. S.C.P., Q.Z., R.D., T.W., K.Z. performed
   methodology; S.C.P., Q.Z. performed validation; S.C.P., Q.Z., R.D.,
   S.K., L.Q., A.A., M.V. performed formal analysis; S.C.P., Q.Z., R.D.,
   J.X., C.P., B.L., K.Z. performed investigation; A.A., M.V., K.Z., Q.Z.,
   R.D., T.W. provided resources; A.M., S.C.P., Q.Z. wrote the original
   draft; A.M., G.H.W., Q.Z., S.C.P. reviewed and edited the manuscript
   with inputs from all authors; Q.Z., S.C.P. performed visualization;
   A.M., G.H.W. performed funding acquisition and supervision. All authors
   reviewed the results and approved the final version of the manuscript.

Supporting information

   Supporting Information
   [320]ADVS-11-2305547-s001.pdf^ (4.5MB, pdf)

Acknowledgements